You have the keyword. Now you need the page. This guide goes from topic to published file: how the dispatch works, where the draft is edited, how a website is connected, what happens to a file that is already there, and where to look a week later to see whether search engines indexed the page and the AI platforms cited it.
Start from the vacant question on your tier
Do not start from what you would like to write. Under the position line on the Brand page, a question on your tier that no AI answer has named you for is a vacant spot. After each collection the AI CMO’s suggestion card is prefilled with that cell, and clicking it is the dispatch. You can also say “write a web page that answers X” in chat; the scope is the same, the three phrasings of that one keyword. Where tiers and vacancies come from is in Climb from the tier you are on.
What the web writer hands back
The web writer writes long-form: sections with headings, and every figure with its source. A draft with a figure that has no source is sent back and rewritten once. If it still has none, those spots are flagged for you to check. Nothing slips through quietly. How to write so that AI answers pick the page up is in Writing for AI answers.
Edit in Solve
Drafts live in the list on the Solve tab. Every save is a version, authored by the writer or by you, and versions can be compared and restored. Editing an existing draft makes a new version of the same piece, not a copy; the writer changing the title is still the same piece. Publishing takes the latest version, so there is nobody to notify when you are done.
Connect the website
Websites are connected under “Websites & accounts” in Settings, in one of two ways:
| Method | What to fill in |
|---|---|
| GitHub Pages | the repository, the content folder, and a token that can write to it |
| Your own server over SSH | user@host, then a password or a private key pasted whole. On first contact you confirm the host fingerprint |
A server the cloud cannot reach, say one on an internal network only, gets a bridge command; run it once on the server and the connection is up. Click the connection’s name to rename it or add a note; the publish panel shows both.
Publish from the draft
Click publish on the draft, pick a website connection, and the AI CMO puts up an approval card: the target path, and whether a file already sits there. An existing file is backed up and then replaced, never overwritten in place. Nothing touches the server until you click “approve and publish”. The final confirmation is always yours.
The published page becomes an asset, listed under “published” in the Assets section of the Brand page. If Bing Webmaster Tools or the Baidu Search Resource Platform is connected, the page is submitted for indexing on its own, and the indexing specialist’s report shows what was submitted and what was accepted.
Collect again a week later
A published page is not a cited page yet. Source agents collect on a schedule, Monday at 9 by default. After the next round, look in two places: the “AI citation sources” block of the brand performance report, for this page; and the “your pages” block of the search traffic report, for impressions. Neither showing anything yet is normal. Wait one more round.
Before you start
- Keywords confirmed, and a vacancy on your tier. If there is none, the suggestion card points at a different cell; take that one.
- The website connection tested. The GitHub token needs write access; the SSH fingerprint is confirmed on first contact.
- Read the target path before approving. The existing file is backed up, but a page published to the wrong path was still approved by you.
Further reading: Writing for AI answers · Real search terms into your keywords